How can superintelligent aliens think?



The subject of superintellectual artificial intelligence does not lack attention from computer scientists, philosophers and the media. Nick Bostrom's recent book is devoted to the development of superintelligence on Earth, but for those who think about extraterrestrial civilizations, this book has many interesting things. The author distinguishes three types of superintelligence:

(1) Fast superintelligence - even a human emulator in principle could work so quickly that a candidate's thesis could be written in one hour.

(2) Collective superintelligence - individuals may not be superintellectual, but the collective of them is superior to human intellect.

(3) Qualitative superintelligence - at least, thinks as quickly as a person thinks, and is much smarter in all spheres of people's activities.

Any of these species can coexist with one or two others.

It is necessary to establish whether these data can be of the same purpose. Here is the opinion of the futurist of Bostrom:

The assertion about orthogonality: "Intellect and basic goals are orthogonal: almost any level of intelligence can in principle be combined with almost any basic goal."

Bostrom emphasizes the fact that in the future many very unexpected kinds of super-intelligent artificial intelligence can arise. He gives a sobering example of superintelligence, whose main purpose is to make staples. At first, the similar goal may seem too harmless and modest, but Bostrom draws attention to the fact that for its implementation, the superintelligence can use all terrestrial forms of matter, while destroying the biological life. Bostrom warns that the superintelligence arising on Earth can be unpredictable and "completely alien" for us.

He talks about several scenarios for the development of the supramental artificial intelligence. For example, the supramental artificial intelligence can appear not because of human intent, but in a way that is incomprehensible to clever programmers. Moreover, Bostromber perfectly seriously considers the probability that the terrestrial superintelligence can be biologically inspired, that is, due to the reengineering of algorithms, which, according to cognitivistics, describe the human brain, or by scanning the contents of the human brain and loading it onto a computer.

Throughout his work, Bostrom emphasizes that since the superintelligence is unpredictable and its actions are difficult to control, it can represent a very serious existential risk for our species, and accordingly, we must always be on our guard. This should greatly cool our fervor and in the context of the search for contact with aliens.

Although difficult to express the main objectives of superintelligence, Bostrom outlines several possible instrumental goals, given that they support any major goal:

The statement about instrumental convergence: "It is possible to establish several instrumental values ​​that are convergent in the sense that achieving them will increase the possibility of realizing the agent's goal for a wide range of basic objectives and a wide range of situations, implying that these tool values ​​will apparently be implemented by a wide spectrum of event-dependent intellectual agents. "

Our New YouTube Video


Please subscribe to our youtube channel BHASKBAN


Video Embed by :Bhaskban

The goals set by Bostrom are the improvement of technology, the acquisition of resources, cognitive development, the integrity and self-preservation of the content of the goals (the future superintelligent personality will become the same goals and pursue). Bostrom emphasizes that self-preservation can include individual or group conservation, and that it may have a secondary significance with respect to the conservation of the species that should serve as artificial intelligence.

For an alien organism with superintelligence, which is based on the reengineering of the alien brain, which includes the loading of the latter, we will use the term "biologically inspired superintelligent alien" or BISI. Although BISI has loaded the brain of a biological species that created a modern superintellectual organism, the algorithms of such an organism can deviate from the algorithms of the biological model at any time.

In the context of understanding the extra-intellectual superintellectuality of a biologically inspired superintelligent alien, they are of particular interest. Since, if Bostrom is right and there are many ways to create a superintelligence, but some extraterrestrial civilizations use to load superintellectuality or other forms of reengineering, it is likely that biologically inspired superintellectual aliens are the most common form of extraterrestrial superintelligence. In fact: many types of superintelligence are possible due to unexpected side effects of programming, which is used by extraterrestrial civilizations. (For example, many artificial intelligence programs on Earth are not based on human brain research). In this regard, the high probability that the class of the supramental artificial intelligence is very heterogeneous and between its representatives the similarity will be very insignificant. It is likely that BISI will be on each other like a lot more than other super-intelligent artificial intelligence. In other words, biologically inspired superintellectual aliens can form a very cohesive subgroup because of the striking dissimilarity of the remaining forms of superintelligence.
If created by a variety of biological species, biologically inspired superintellectual aliens are scattered throughout the galaxy, what are the common properties of representatives of this class? biologically inspired superintelligent aliens have a specific origin, and here it is worth highlighting two points on the basis of which common cognitive abilities and goals can be formed:

(1) BISI comes from beings who have food needs, safety, reproduction, cooperation, competition, and so on.

(2) The life forms on the basis of which BISI is created evolved to overcome such biological limitations as the slow processing speed of information and the spatial boundaries of the incarnation.
Is it possible, based on (1) or (2), to formulate features that are common to the members of many superintellectual extraterrestrial civilizations? It seems to me that yes.

Consider the following moment (1). Intellectual biological life tends, first of all, to reproduction and self-preservation, so that, in all likelihood, the main goals of BISI will be reproduction and self-preservation, or at least self-preservation and reproduction of members of the superintellectual society. If biologically inspired superintellectual aliens are interested in reproduction, we, considering their colossal computing resources, can expect that they create artificial universes with artificial life and even intellect or superintelligence. If the purpose of these creatures is to be "children", they may retain the goals listed in paragraph 1.

I can be objected here: they say, it's quite useless to talk about the forms of BISI, because they have the ability to change their basic architecture in a variety of ways and unpredictably. At the same time, any biologically inspired motive risks being reprogrammed and significantly reduced. But here, it seems, there are boundaries. If the superintelligence has a biological basis, self-preservation can be its main goal. In this case, he may consider the best, without encroaching on the basics of his architecture, to make any improvements to it. He can think about this way: "rebuilding my architecture radically, I will cease to be myself." For example, downloads can be implemented so as not to damage the properties of intelligence that are very important for its biologically inspired existence.

Now let's move on to the second moment. The self-improving superintelligence and superintelligence developers can move away from the original biological model in various and unpredictable ways, but, as I noted, biologically inspired superintelligent aliens may not want to radically rebuild their architecture. Let's try to establish useful for him cognitive abilities, that is, those abilities of sophisticated forms of biological intelligence, without which, apparently, these forms are not able to exist and due to which the superintelligent can fulfill its main and instrumental goals. In addition, we will try to establish those properties that are unlikely to arise in BISI, although they do not call into question the fulfillment of its goals.

If it's true what the second point says, we can expect, for example, the following:

1. To understand the models of thinking of biologically inspired superintellectual aliens, one can come by studying the computational structure of the brain of a biological species that created this superintellectual organism. In the cognitive sciences, one of the most important means of understanding the computational structure of the brain is "connecticomics" - the area of ​​study of the connection map or the neuroelectric circuit of the brain. Although the connectors of this BISI and its creators may not coincide, it is quite possible that some functional and structural connections in both cases will be the same. In this case, you can establish interesting deviations of artificial intelligence from the original.

2. BISI can have views that do not depend on the point of view. At a high level of information processing, our brain creates internal representations about objects and people, not depending on the point of view. Let's remember how we approach our front door. We have done this hundreds of times, and maybe thousands, but, in terms of building the movement, each time in different ways, since we have never moved in the same way. And never once did things appear in front of us at exactly the same angles of view. But at a relatively high level of information processing, our brain creates such mental representations that do not depend on the point of view. Apparently, the ability to create such views is necessary for the development of any intelligence with a biological basis, since categorization and prediction can not be performed without them. The mobile system needs tools to identify items that are in an ever-changing environment, and therefore there are stable internal representations. Now it should be clear why one should expect the presence of such representations in artificial biological systems on a biological basis. With what statiobiologically inspired superintelligent aliens will abandon object-invariant representations, if it is a mobile system or has mobile devices for remote communication?

3. BISI will have linguistic mental representations, and they will be combinatorial and recursive. It is worth paying attention to the fact that human thought has such a comprehensive and key property as combinatoriality. Let's take the following proposition: Italian wine is better than Chinese wine. Probably, before this thought never occurred to you, but you could always understand it. The fact is that thoughts are combinatorial, because they are composed of familiar parts and are united according to known rules. The rules apply to constructions from primitive components, which are themselves grammatically built, and to the primitive components themselves. Grammatical mental operations are very useful: the combinatorial nature of thinking makes it possible to understand and add sentences using grammar and atomic components (such as, for example, "wine", "Chinese"). So the idea is productive: thanks to combinatorial syntax, in principle, you can, having fun, create an infinite number of different representations.

The brain needs combinatorial representations for the reason that potentially there are infinitely many linguistic representations, they require an infinite space for their storage, and the brain does not have an infinitely large storehouse. Even a superintellectual system is useful for combinatoriality. Possessing the ability to have huge computing resources, which in many cases allow us to select suitable proposals from oral storage for verbal or written statements from the repository, this system is unlikely to cancel such a wonderful property of biological intelligence as combinatoriality. Otherwise, it will turn out worse than it was, because, no matter how huge the storage, it is always finite, and, therefore, there may not be some proposal at the given moment in it.

4. A BISI can have one or more global workspaces. When we look for a fact or focus on something our attention, our brain provides access to sensory or cognitive data to the "global workspace," where this information for more focused processing is transmitted to the attention and working memory systems, as well as massive parallel channels in the brain . The global work area functions as a single space in which important information coming from the senses is examined at the same time, so that the thinking being can make carefully thought out decisions and act intelligently with all the facts at his disposal. In general, it would be inefficient to have the ability to feel or cognize, not integrated with other abilities, because sensory or cognitive information could not figure in predictions and plans based on the evaluation of all available information.

5. Mental processing of BISD information can be understood with the help of functional decomposition. In order to understand such complex systems as extraterrestrial superintelligence, we have at our disposal a method of functional decomposition. A key feature of computational approaches to brain research is the explanation of cognitive and perceptual abilities by decomposing a particular ability into its causally-organized components, which, in turn, are interpreted in terms of causal organization of their parts. This is the aforementioned "method of functional decomposition" - the key method of explanation in cognitive sciences. It is difficult to imagine a complex thinking machine without a program consisting of causally interrelated elements, each of which consists of causally organized elements.

With all this, superintellectual creatures are by definition creatures that excel in any field of human activity. Although such a creature can process information using methods that we are still able to comprehend in the main features, it is possible to meet with some kind of superintelligence whose calculations will be much beyond the human ability to understand. As noted, I consider only the scenario according to which the intellectual work of the supramental artificial intelligence can be reached to our understanding. At the same time, the key to understanding the main moments of the complex mental life of some biologically inspired superintellectual aliens gives the achievements of cognitive sciences. Next, we need to consider several important social and philosophical consequences of the discovery or creation of a superintelligence. And we will begin with a discussion of the problem of control in the framework of the discussion on the active search for extraterrestrial intelligence.

Powered by Blogger.